You are here

Mac Processor Upgrades For Music

Speed Tests By Mike Watkinson
Published September 2003

Mac Processor Upgrades For Music

Processor upgrades are a great idea for making older computers in general household use last longer. But how well does a processor speed boost translate into improved performance in musical applications on older Macintoshes? We find out...

Despite the extensive testing I carried out on processor upgrades for iMacs, Power Mac clones and Beige G3s in SOS December 2001, I was unable to form a positive conclusion either for or against the concept. As processor speeds have edged gently upwards, upgrades have followed, usually a few hundred Megahertz behind. At the time of the tests carried out for this article, earlier this year, Powerlogix had a range of upgrades featuring 800MHz and 1GHz processors, and Apple's fastest machine was 1.25GHz. But as this article goes to press in late July, the situation has changed; Powerlogix are now supplying upgrades at up to 1.2GHz, and Apple's fastest G4 processor is 1.42GHz. The 800MHz Powerlogix upgrades tested in this article are still available, but their prices have dropped between testing and publication (see the 'Upgrades & Pricing' box towards the end of this article).

Although Apple do seem to have a policy of nudging down Power Mac prices with each speed bump, there are of course many people using older machines who cannot justify the expense of a new machine, especially if a simple and cheap upgrade could achieve the performance they require. With this in mind, I decided to stage a rematch, with newer machines, and with upgrades which have a tested clock speed nearly double that of the processors they will replace.

The Machines & Upgrades

Although both of the machines I used in my first set of tests this time could be said to belong to the most recent family of Power Mac (those based around G4 processors), both had processors running at less than half the current maximum. They are the 450MP (dual-processor) Tower and the 450 Cube (see the 'Test Spec' box opposite for the full spec of the machines used in the tests). Both machines are significant milestones in Power Mac development, the former because it was the first dual-processor machine available. In its unmodified guise, it still feels as fast as current machines when running day-to-day tasks. It is also a timely reminder of how quiet the first Power Macs were compared to today's Dyson impersonators! This particular model is notable in that the hard drive can be plainly heard above the noise of the fans...

The Dual 800MHz Powerlogix upgrade card used in the 450MHz Power Mac G4 tower.The Dual 800MHz Powerlogix upgrade card used in the 450MHz Power Mac G4 tower.The latter machine, the Cube, might have been Apple's greatest achievement. All that power and no fans! It's a shame that Firewire had not, at the time, caught on in the peripherals market in the way it has now, as this is the only high-speed interface the Cube has to offer. In the light of the concerns raised by the audio community over noise issues, a silent computer that supports a range of monitors would surely be a massive hit with those currently forced to hide their CPU under the stairs. Upgrading a Cube, therefore, might give the ultimate combination of subtlety and power for those who run a compact project studio. My initial tests were not encouraging; it seemed the power button had gone west, and I found reference to a fault of this kind at http://news.com.com/2102-1040-248678.html?legacy=cnet. However, in this case, the problem turned out to be nothing more serious than a loose wire dislodged by a previous reviewer!

Powerlogix, who supplied both of the upgrades on test via their UK distributor, AM Micro, have a very comprehensive web site which shows exactly which processor upgrade is compatible with each Apple product. The 450MP machine has a 100MHz system buss and AGP graphics support, so I chose the 800MHz PowerForce Dual G4 Series 100 upgrade. This card carries two 7450-type Motorola processors with L2 and L3 cache.

In the box along with the card (attached to its heatsink) was a heatsink fan and power cable, two bolts to attach the fan, an installation CD, and some very detailed installation instructions (see the box on this on the next page).

The Cube should theoretically be able to accept a dual-processor upgrade, but Powerlogix are working to fix the heat-dissipation issues, which currently mean that only single-processor options are supplied for this machine. I chose the 800MHz PowerForce Single G4 Series 100 Cube upgrade, incorporating one 7450-type processor and identical cache specifications to the Dual model.

Installation

Installing the upgrade in the 450MHz G4 Tower was easy. First, the old processor is unscrewed...Installing the upgrade in the 450MHz G4 Tower was easy. First, the old processor is unscrewed...... and removed from its connector.... and removed from its connector.Then the new processor, with its heatsink and fan, is screwed in its place.Then the new processor, with its heatsink and fan, is screwed in its place.Installing the other upgrade in the G4 Cube was much harder. It comes apart OK...Installing the other upgrade in the G4 Cube was much harder. It comes apart OK...... but several fiddly circuit boards have to be removed inside...... but several fiddly circuit boards have to be removed inside...

... before you can even get the old processor off and put in the new fan. The 40-step installation guide is a must, as are the 24 pictures that accompany it!... before you can even get the old processor off and put in the new fan. The 40-step installation guide is a must, as are the 24 pictures that accompany it!Installation into the Power Mac 450MP was very simple. The case has the latching side door still used in current G4 machines, and the processor sits on the motherboard which lies flat and unencumbered when the door is fully open. The old processor simply lifts off its connector once the retaining screws have been removed. The upgrade is positioned by lining up the screw holes in the heatsink with the holes in the motherboard, and once this is done, the connector snaps (gently!) into place, and the upgrade is secured with the original screws. The cooling fan is secured on top of the heatsink using the two bolts provided, though you must make sure that the airflow indicator is pointing down towards the heatsink (see picture, below right). The last task is to connect the fan's power cable to a spare power supply socket; then you simply close the case and you are ready to boot up.

Physical installation in the Cube is a lot more tricky (see the bottom row of pictures below). Although I tend to enjoy this sort of thing (it brings back many a happy memory of rainy days whiled away with nothing more than an Airfix kit for company), I can imagine that most musicians might be a little apprehensive about dismantling the innards of such a carefully designed and compact set of electronics. If the thought of 40 steps of instructions fills you with even a small amount of dread, I would thoroughly recommend the employment of an authorised Apple service engineer to carry out the installation. 24 photos also accompany the instructions, but if this is still not enough detail for you, Powerlogix also include an installation movie on the software CD!

The kit also contained a fan and power cables, connecting screws, installation CD and instructions. Did I say fan? Yes, although the G4 Cube is a fan-less design cooled by convection alone, Powerlogix recommend the installation of one due to the extra heat generated by the faster processor. Surprisingly there is a space (with guides) for a fan inside the Cube's 'core' — it's almost as if the designers were expecting one to be fitted.

Finally, both upgrades require a software installation from the CD provided. This places an extension in the System folder that enables the L3 cache for operation under OS 9. However, this is not necessary under OS X.

The Tests

In the December 2001 SOS processor upgrades article, I carried out various types of test involving software 'benchmarks', file-rendering and plug-in counts. My conclusion at the time was that software benchmarks, while useful as rough guides, do not give a sufficiently accurate prediction of what will happen when using real applications. One area of debate was the non-linearity with which the number of Logic plug-ins available relates to processor speed compared with VST plug-ins.

Table 3: All machines (OS X plug-in counts at 48kHz).Table 3: All machines (OS X plug-in counts at 48kHz).Nearly two years later, and given the advent of OS X, I was keen to check the difference in performance between equivalent plug-ins on OS 9 and OS X, as well as the relative gains due to processor upgrade. In OS 9, in Logic, I used two Logic-native plug-ins (stereo and mono Platinumverbs) and two audio Instruments (EVB3 and ESP), as well as two Waves VST plug-ins (the C4 Stereo compressor and RVerb reverb, both known for their heavy processor consumption). I also tested the two Waves plug-ins in Nuendo. I did not measure VST instruments in Nuendo (or Cubase SX in OS X, for that matter) since both of these hosts support a maximum of eight, which rather defeats the object of the test!

I used the same plug-ins and Instruments in Logic under OS X (with the exception of the Waves C4 and RVerb, since at the time of testing, these were VST-only), and ran the Waves plug-ins in Cubase SX, which has an audio engine with a very similar feel and setup to Nuendo, since version 2 for OS X was not quite available at the time of testing.

Note that there are no screengrabs in this article from Mac OS X. This is because the screen-capture function in OS X loads the Mac's processor to the extent that previously running maximum plug-in counts fail! OS 9's screenshot function does not have this effect.

To keep the playing field as level as possible, both operating systems were installed from scratch on newly formatted system disks. In OS 9, the default extension set was left enabled and Virtual memory was off. The disk cache was left set to default, and MOTU's 896 (the audio interface used due to the Firewire requirements of the Cube) was set to 1024 samples per buffer. Logic's control panel for the Audio Hardware driver was set to 'Process Buffer Large', with the Larger Disk Buffer not enabled. The equivalent disk buffer settings in Cubase SX and Nuendo (both use the same setup method) were left at default, which is four buffers of 128K. In OS 9, each host was assigned 300MB of memory (this is, of course, not a user option in OS X).

Logic requires an audio event to be playing through an audio channel before a plug-in loads the processor, so a new event (a copy of the first) was added each time the full number of inserts (15 per channel) was reached.

Nuendo and Cubase SX don't require an audio event to be playing (or even to be present) for plug-ins to load the processor, which can be a nuisance if you are trying to edit a track close to the limit, since the most noticeable sign is that the graphics slow right down. However, to try and keep comparisons fair, I added a new event each time an audio channel reached its maximum quota of insert plug-ins (eight for both of these hosts).

When the maximum number of plug-ins is reached, Logic simply stops playing and flags up a warning telling you that you have run out of power. Nuendo/Cubase SX respond differently; the audio starts to break up (with gaps in playback, and/or a grainy, distorted sound), despite the fact that sometimes another plug-in can be added before the sequencer stops playing altogether (with no warning message).

It is therefore a little more subjective as to when the processor is pushed beyond its capabilties with these applications. I decided that the cutoff point for the sake of these tests was when I detected any break-up in the sound, which proved a particular problem with the Waves plug-ins due to their heavy CPU consumption.

In order to create a further point of comparison, I took test results at 48kHz and 96kHz sample rates on the 450MP machine (before its processor was upgraded). The majority of the tests were carried out at 48kHz. For reference purposes, the same tests were run on a dual-1.25GHz Mirror Door machine.

Host Plug-In type Plug-In name OS X At 48KHz At 96kHz OS X At 48KHz At 96kHz
Logic v5.5.0 Logic Platinumverbs Mono 16 7 15 7
Platinumverbs Stereo 12 5 10 4
EVB3 Stereo 3 1 4 1
ESP Stereo 17 7 15 7
Waves C4 Stereo 7 3
Rverb Stereo 4 2
Nuendo v1.6 (OS 9)/Cubase SX (OS X) Waves C4 Stereo 8 3 4 2
Rverb Stereo 3 1 2 1

Host Plug-In type Plug-In name 450MP 450MP Cube 450 Cube Mirror Door
(no upgrade) (800MP upgrade) (no upgrade) (800SP upgrade) Dual 1.25GHz
Logic v5.5.0 Logic Platinumverbs Mono 16 20 14 17 36
Platinumverbs Stereo 12 16 10 14 27
EVB3 Stereo 3 5 3 4 9
ESP Stereo 17 22 15 20 38
Waves C4 Stereo 7 11 7 9 17
Rverb Stereo 4 6 3 5 10
Nuendo v1.6 Waves C4 Stereo 8 12 6 9 15
Rverb Stereo 3 4 3 4 9
Host Plug-In type Plug-In name 450MP 450MP Cube 450 Cube Mirror Door
(no upgrade) (800MP upgrade) (no upgrade) (800SP upgrade) Dual 1.25GHz
Logic v5.5.0 Logic Platinumverbs Mono 15 19 13 16 35
Platinumverbs Stereo 10 13 9 12 24
EVB3 Stereo 4 5 2 4 8
ESP Stereo 15 20 12 17 36
Waves C4 Stereo
Rverb Stereo
Cubase SX v1.5.1 Waves C4 Stereo 4 6 6 7 12
Rverb Stereo 2 2 2 3 8

Results

To restrict the amount of information presented in each table and allow meaningful comparison, one potentially variable parameter is kept at a constant.

TABLE 1

In Table 1 (shown on the next page), comparison is restricted to one machine in its non-upgraded form, allowing for the following comparisons to be made:

  • A comparison between 48kHz and 96kHz operation under OS 9

Without exception, plug-in counts at 96kHz are less than half that for 48kHz, proving that running at the higher rate increases processor demand by slightly more than the expected factor of two.

  • A comparison between 48kHz and 96kHz operation under OS X

Testing instrument plug-ins: Emagic's EVB3 under Logic in OS 9.Testing instrument plug-ins: Emagic's EVB3 under Logic in OS 9.The situation outlined above is repeated here, with the more processor-intensive Logic plug-ins faring even worse than under OS 9 with the move to 96kHz. Percentage losses appear greater for processor-hungry plug-ins, because it is harder to make small incremental increases near the limit.

  • A comparison between OS 9 and OS X at 48kHz and at 96kHz

Without exception, OS 9 plug-ins perform slightly better than their OS X counterparts. Waves VST plug-ins fare the worst by comparison here, perhaps showing up a loss of efficiency when a plug-in is simply ported to work under OS X, as opposed to rewriting it to operate under a new protocol supported at system level (Logic's plug-ins use the Audio Units protocol under OS X).

  • A comparison between Logic and Nuendo as VST hosts under OS 9

There's no clear winner here! Detractors of Logic as a VST host under OS 9 should take note...

TABLE 2

This restricts the operating system to OS 9, allowing the following comparisons to be made:

  • A comparison between performance gains according to Logic's plug-ins

Logic's plug-ins gain from the processor upgrades by an amount roughly in the range 25-35 percent. It has been shown before that Logic plug-in availability does not increase predictably with processor speed, suggesting that they are influenced by other factors, such as system buss speed. Interpolating in linear fashion between figures for the non-upgraded 450 MP and the 1.25GHz Dual DDR machines could give rise to one of two conclusions. On the one hand, if the 800MHz upgrade leads to a reasonable increase in performance, then the 1.25GHz performs better than expected. On the other hand, if the 1.25GHz performance fulfils expectations, then the 800MHz upgrade disappoints.

  • A comparison between performance gains according to VST plug-ins (and between VST hosts)

Testing more processors: Waves' C4 running under Nuendo in OS 9.Testing more processors: Waves' C4 running under Nuendo in OS 9.Gains with VST plug-ins do not appear to be significantly different to those exhibited by Logic plug-ins. The 1.25GHz Dual machine again exceeds the expectations set up by upgrading from 450 to 800MHz, showing that VST plug-ins also rely on more than just the speed of the processor.

TABLE 3

This restricts the operating system to OS X, allowing comparisons to be drawn as for Table 2. With every figure in this table coming at slightly under that for the comparable figure under OS 9, no evidence is found which conflicts with the conclusions drawn for Table 2. As a whole, this table illustrates that plug-in count is lower for OS X than for OS 9 for comparable hosts.

Both Table 2 and 3 offer an insight into the benefits of multi-processing. Under OS 9, multi-processing is not supported in the OS itself but in the application. Logic purports to assign MIDI duties to one processor and Audio duties to the other. This is backed up by the graphic display of System performance when running many plug-ins. In OS X, it is the OS itself that supports multi-processing, allowing each host to gain the benefit of both processors at all times and for all duties. Logic's performance monitor accordingly has only one meter for processor loading in OS X, where two are shown in OS 9.

The figures, however, show only slight gains between the single-processor Cube and its dual-processor equivalent, in both the original and upgraded forms, and these differences are not exaggerated as you might expect by the move from OS 9 to OS X. This shows that in the case of plug-in counts at least, either Logic's OS 9 workaround for multi-processing was not the bodge everyone claimed it was, or, more likely, that proper multi-processing is still not available to these hosts under OS 10.2.3.

My guess is that there is some truth in both of these positions!

Upgrades & Pricing

PFD 800100: £616.88

  • Dual 800MHz Powerlogix upgrade for AGP dual-processor G4 Power Mac with 100MHz system buss.
  • Two 800 MHz 7455 processors.
  • L2 cache: 256K on-chip L2 cache.
  • L3 cache: 2MB per processor.
  • 100MHz System buss.

PFG4 800S233C: £245.58

  • Single 800MHz Powerlogix upgrade for Power Mac G4 Cube.
  • One 800MHz 7455 processor.
  • L2 cache: 256K on-chip L2 cache.
  • L3 cache: 2MB.
  • 100MHz System buss.

Prices include VAT and are correct at time of going to press (July 2003).

AM Micro +44 (0)1392 426473.

+44 (0)1392 439346.

info@ammicro.co.uk

www.ammicro.co.uk

Conclusions

Deciding whether or not to upgrade your processor is a tough decision based on your needs and financial position. To play devil's advocate for a moment, the advantages of ignoring the processor-upgrade concept altogether and putting your money towards one of Apple's latest machines are numerous — particularly if time hangs on your hands as well as money, and you can wait until the recently announced super-powerful G5s are available. A new machine will afford you significant performance gains — and don't forget, this will be in all areas, not just with regard to the number of processing and instrument plug-ins available to you.

Of course, if you simply cannot afford to make the leap to a new machine, then upgrading is your only alternative. However, depending on what you use your Mac for and the reason why you feel compelled to upgrade, this may or may not be worth your while. Test figures for all manner of general-purpose applications are available from sites such as www.xlr8yourmac.com and www.barefeats.com if you wish to consider an upgrade to enhance your household Mac useage, but if increased audio performance is the only reason for the upgrade, then it might be wiser to wait, save and look at buying a new machine. On the basis of my findings in these tests, although there are increases in performance to be had from your favourite audio application, the percentage gains are not as large as you might expect from the numerical specifications of the upgrades.

Published September 2003